Artificial intelligence
From RPA to intelligent automation: synergy between bots and AI
Publiée le November 18, 2025
Publiée le November 18, 2025
Traditional RPA is limited to repetitive tasks based on explicit rules. It cannot interpret unstructured data or make complex decisions. What’s more, the slightest variation in an application’s interface can break a robot if it isn’t programmed to handle it. To go one step further, editors have integratedartificial intelligence (AI) technologies: OCR, machine learning, natural language processing (NLP) and, more recently, large-scale language models (LLM). This hybridization is at the heart ofhyper-automation and paves the way for the advent ofautonomous agents.
Vision and OCR: optical recognition (OCR) and computer vision engines enable scanned documents or invoices to be read. UiPath Document Understanding combines OCR and supervised learning to extract structured data from semi-structured documents (invoices, contracts) and integrate them into an ERP system.
Natural language processing: NLP can be used to understand e-mails, categorize requests and extract intentions. For example, an insurer can automate the sorting of incoming e-mails (claims, complaints, general questions); an AI model classifies the messages and triggers RPA robots that create dossiers or send standard responses. Agentic automation solutions such as those from UiPath use generative models to enable AI agents to write responses and summarize documents.
Machine learning and predictive analytics: ML algorithms identify patterns in data and make predictions (risk scoring, fraud detection). AI can decide to automatically approve a case when certain criteria are met, or to send a case to a human agent for verification.
LLM and generative AI: new-generation language models (GPT, Gemini, Claude…) can generate text, write code or explain processes. UiPath is betting on agents capable of generating automation scripts from a simple natural language instruction.
Classification and automatic response to e-mails: an AI system analyzes the content of the message, identifies the intention (request for quote, complaint, current question), then transmits the data to an RPA robot which fills in the customer file and sends a personalized response.
Intelligent invoice processing: OCR extracts amount, date and supplier; ML model detects anomalies; RPA matches invoice to purchase order and triggers payment. According to AutomationEdge, this combination of NLP and OCR enables claims to be processed 75% faster than humans in the insurance sector (automationedge.com).
Virtual assistants and chatbots: LLM-based conversational agents act as user interfaces, retrieving the necessary information and calling bots to perform actions (book a flight, update an order). UiPath refers to “agents” capable of interacting with users and connecting to multiple systems to act on their behalf.
Automated decision-making: by coupling a rules engine with a predictive model, RPA can automate the granting of credit or insurance for simple cases, leaving only complex cases to human analysts. This approach reduces processing times and standardizes decisions.
Introducing AI into RPA poses several challenges:
Data quality: to produce reliable models, organizations need to control the quality and governance of their data. Bias or erroneous information can lead to unfair or incorrect decisions.
Explicability: it is sometimes difficult to understand why an AI model has made a particular decision. Transparency and auditing mechanisms are needed, especially in regulated sectors such as banking and insurance. Platforms include explainability and human-in-the-loop modules to validate certain decisions.
Security and confidentiality: AI requires access to large amounts of data. We need to guarantee the protection of sensitive information and comply with regulations (RGPD). Robots and models must be deployed in secure environments, with fine-grained management of access rights.
Skills: implementing intelligent automation requires skills in data science, RPA development and governance. RPA Centers of Excellence (CoEs) are evolving into “intelligent automation CoEs”, integrating data scientists, AI specialists and business experts.
Agentic automation is the next logical step in hyper-automation. It involves building autonomous agents that orchestrate robots, AI and humans to achieve a goal. An agent manages the sequence of actions (collecting data, executing RPA scripts, soliciting human input) and adjusts its behavior according to results and feedback.
UiPath, Blue Prism and Microsoft are investing heavily in this vision. In its July 2025 blog post, UiPath explains that automation is set to evolve towards agents capable of making decisions, adapting to change and collaborating with users. The platform already integrates self-healing functions and proposes a roadmap for LLM-based agents.
Intelligent automation represents a major evolution from conventional RPA. By integrating AI, companies can automate more complex processes, process unstructured data and offer a more natural interaction to their customers. However, this evolution requires new skills, strengthened governance and a heightened awareness of ethical issues. The organizations that succeed in this transformation will be those that approach automation as a comprehensive program, blending RPA, AI, data culture and change management.